Research On Signal Detection Method Based On Spectrum Characteristic And Deep Learning

With the development of the times,radio communication has posed great impact on our life.The urban electromagnetic environment is increasingly complex.Many illegal signals such as pseudo base stations,black broadcasting,and interference signals abuse spectrum resources,which seriously affect daily production and life,and bring challenges to radio management.Signal detection is the core technology of radio management,and the research of signal detection methods is of great significance to radio management.The traditional signal detection method has the advantages of good real-time performance and convenient engineering implementation.However its detection performance is sensitive to threshold selection,so how to select the threshold in complex electromagnetic environment becomes a traditional signal detection problem.Therefore,this paper introduces a deep learning method widely used in the field of image target detection in recent years,which can automatically recognizes the frequency domain characteristics of the signal by the neural network,without artificially setting the detection threshold.In this paper,the convolutional neural network signal detection algorithm based on short-time Fourier transform is studied.The algorithm extracts time-frequency map features by signal short-time Fourier transform,and uses convolutional neural network to construct signal detection model,frequency position detection model and respectively modulation mode identification model.In this paper,the experimental platform is used to collect a large number of real data with different bandwidth,signal-to-noise ratio,modulation mode and frequency position to complete the model training and algorithm performance test verification.The results show that the performance of the proposed algorithm is similar to that of the traditional signal detection method.Without being constrained by the threshold,the proposed algorithm has higher practical value in the complex electromagnetic environment.This paper presents a neural network signal detection algorithm based on modified periodogram.The algorithm obtains one-dimensional spectrum features by correcting the periodogram calculation,reduces the random fluctuation of spectrum data by morphological filtering,and constructs the signal detection model by using convolutional neural network.In this paper,a large amount of real data is used to analyze the performance of the model,and the influence of key parameters of the model on its performance is studied.Aiming at the multi-target signal detection requirements,this paper proposes a multi-target signal detection algorithm based on sliding window,which increases the detection time,but can differentiate multiple target signals with different frequency positions,and significantly reduces the error of the detection of signal frequency.